LR-COBRAS: A logic reasoning-driven interactive medical image data annotation algorithm.

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Healthcare
• Computer-Aided Diagnosis

Background:

• Increasing volume of medical imaging data necessitates efficient annotation.
• Manual annotation is time-consuming, error-prone, and costly.
• Deep learning models require large, accurate datasets, posing a challenge for annotation.

Purpose of the Study:

• Introduce LR-COBRAS, an interactive computer-aided algorithm for medical image annotation.
• Improve precision and efficiency in medical image annotation tasks for healthcare professionals.
• Optimize the creation of training datasets for deep learning in medical imaging.

Main Methods:

• LR-COBRAS utilizes a logic reasoning module to enhance must-link and cannot-link constraints.
• The algorithm automatically generates constraint relationships, minimizing user interaction.
• Employs rules like symmetry, transitivity, and consistency for balanced automation and clinical relevance.

Main Results:

• LR-COBRAS demonstrated superior clustering accuracy and efficiency compared to existing methods.
• The algorithm significantly reduced the interactive burden on users.
• Experiments on MedMNIST+ and ChestX-ray8 datasets confirmed robustness and applicability.

Conclusions:

• LR-COBRAS offers a novel, intelligent solution for medical image annotation.
• The interactive approach enhances accuracy and efficiency in dataset creation.
• This algorithm supports the development of more stable and reliable deep learning models for medical image analysis.